Top 2% on Product Hunt, Top 3 by Comment Count: This AI Workflow Played a Key Role

When Draft made it to the Product Hunt Feature page, I didn’t have a hunter, and I didn’t have an early user base.

Looking back, the Product Hunt Gallery was one of the few key things I did right.

This article is not about design theory. It is about how someone without professional design training used AI to create a usable set of product showcase images from scratch.

How can a small team, with very limited resources, use an AI-native approach to break down something that normally requires design, copywriting, product, and marketing collaboration into a workflow that can be executed, reviewed, and reused?

In this post, I want to share my practical experience:

How I went from zero to creating a set of marketing images that played a key role in our Product Hunt launch.

One key understanding

Before making any images, you need to understand one thing:

On Product Hunt, users are not there to admire your design.

They are making a quick judgment within seconds:

Does this have anything to do with me?
Is it worth clicking into?
Is it worth upvoting?

So, a Product Hunt Gallery is not an image collection. It is a sales deck.

The first hero image is responsible for making people stop.

The following feature images are responsible for answering users’ questions step by step.

Step 1: Filter for the features that are truly valuable

When you have been building a product for a while, it is easy to fall into a trap:

Since we built so many features, shouldn’t we show all of them?

Remember: your gallery is not a product manual.

Users will not carefully study every feature. They are deciding within a few seconds whether this product is worth continuing to look at.

So, the features that deserve to be shown are not necessarily the ones you spent the most time building.

They are the ones that best help users understand the product’s value.

My filtering criteria were:

Can this feature help users understand what the product is faster?
Can this feature show product differentiation?
Can this feature be expressed visually at a glance?
Does this feature map to a real pain point?

Take Draft as an example. Many hardcore features did not make it into the gallery.

Not because they were unimportant, but because their understanding cost was higher than the target user’s cognitive load.

For first-time users, “turning AI conversations into organized, reusable knowledge” is much easier to understand than “supporting multiple rich formats.”

:::toggle{level=3} Practical Prompt: Filter features suitable for the Gallery Use case:

When you do not know which features should go into your gallery, first list all your features. Let AI help you do the first round of filtering. But the final decision must still be yours.

Prompt:

I am an indie developer preparing to launch on Product Hunt.

My product is:
[One-sentence description of your product]

All of my product features are:
[List all features]

Please do not treat this as a feature list.
Treat it as a feature selection task for a Product Hunt gallery.

Please help me choose the 3–4 features that are most suitable for the gallery.

Selection criteria:
1. Can users understand the product within a few seconds?
2. Does it show clear differentiation?
3. Can it be expressed visually?
4. Does it address a real pain point?
5. Can it encourage users to keep viewing the next image?

Please output:
- Features recommended for the gallery
- Features not recommended for the gallery
- The reason why each feature is selected or rejected

:::

One reminder here:

AI can help you filter, but the final decision must be yours.

Because AI does not understand better than you what your product should be remembered for at this moment.

Step 2: Be disciplined. Each image should express only one key point.

The easiest mistake to make when creating a gallery is trying to say too much in every image.

The title wants to explain positioning.

The subtitle wants to explain the feature.

The screenshot is packed with details.

Then two more selling points are added in the corners.

It looks like a lot of effort, but users remember nothing.

So before making an image, you should first think clearly:

What exactly is this image trying to communicate?

Ask yourself a few questions:

If this image can only express one key point, what is it?
After seeing this image, what sentence should users remember?
What is this image trying to convince users to believe?
If this image were removed, what would users fail to understand?

If you cannot clearly define the goal of the image yourself, AI will probably not be able to draw it clearly either.

For example, one image can be responsible for saying:

Publish once, share anywhere — via Link or Card.

Each image should communicate only one key point. That is how users remember it.

:::toggle{level=3} Practical Prompt: Define the communication task for a single image Use case:

You have already selected one feature, but you are not sure how the image should express it.

The purpose is not to generate the image immediately. The purpose is to first clarify “what job this image needs to do.”

Prompt:

I want to create a Product Hunt feature image.

This feature is: [feature name]

Please do not create the image yet.
Instead, help me define the communication task of this image.

Please output:

1. What is the core message of this image?
2. What sentence should users remember after seeing it?
3. What should this image convince users to believe?
4. If this image were removed, what would users fail to understand?
5. Recommended title and subtitle
6. Recommended visual layout

:::

Very often, the issue is not that we do not know how to use AI.

It is that we have not clearly explained the task.

The clearer the task, the more AI behaves like a collaborator. The vaguer the task, the more AI behaves like a random generator.

Step 3: Find references and use AI to break down the style

As someone without a professional design background, my method was simple and rough:

Go to Product Hunt and save gallery images from strong products
Use AI to break down their style, layout, and color palette
Ask AI to adapt those references based on my product’s tone

The point of this step is not to copy other people’s images.

It is to translate “I think this looks good” into language AI can understand.

For Draft, I referenced the layout of a specific image.

Then AI created a version based on that visual structure.

A few baseline rules for keeping the gallery polished:

The gallery should maintain a consistent color tone
Pure color backgrounds can feel boring, so add subtle texture
Try not to repeat the exact same layout in every image
Fonts, colors, labels, rounded corners, and shadows should stay consistent

The polished feeling of a gallery essentially comes from consistency.

A set of images should look like they came from the same product, not like five random images generated by AI.

Step 4: Work backwards. Make feature images first, then the hero image.

The hero image is the most abstract and the hardest to make.

It needs to let users know what the product is, communicate value, and still be attractive.

If you start directly with the hero image, AI’s understanding of your product is still shallow.

It will likely give you something that “looks nice” but does not really say anything.

My trick was:

Make the feature images first, then come back to make the hero image.

After you finish several feature images, AI has already developed a better understanding of your product workflow, core selling points, and visual style.

At that point, its hero image suggestions are usually much more focused.

:::toggle{level=3} Practical Prompt: Reverse-engineer the Hero image from Feature images Use case:

You have already completed several feature images and want AI to summarize the core value and visual theme of the whole gallery before continuing to make the hero image.

Summary prompt:

I have completed several Product Hunt feature images.

Based on these feature images, please help me summarize:

1. What is the core value of my product?
2. What is the easiest sentence for users to remember after viewing them?
3. What is the shared visual theme across the whole gallery?
4. What should the hero image emphasize the most?
5. What should the hero image avoid repeating?

Generation prompt:

Based on the feature images I have already completed, please help me design a Product Hunt hero image.

Requirements:
1. The hero image should feel like the cover of the whole gallery
2. Clearly communicate what the product is in one sentence
3. Do not repeat the detailed content already covered in the feature images
4. Put more emphasis on overall value, first impression, and click-through appeal
5. Keep the visual style consistent with the feature images

Please output:
- Hero image title
- Subtitle
- Recommended visual layout
- Core elements that should appear
- Redundant information that should not appear

:::

This order helped me reduce a lot of rework.

It was not because AI got everything right in one shot. It was because the backward process had already clarified the product understanding, visual style, and key messages.

Step 5: Let a second AI act as the reviewer

The most dangerous moment when making images is when you have looked at something for too long and start thinking:

“Maybe this is good enough.”

So I used another AI as a third-party reviewer.

My workflow was:

GPT generates the image → Gemini critiques the image → I judge the direction and make the final decision

The key is not which model is stronger.

The key is not to let the same AI be both creator and reviewer.

:::toggle{level=3} Practical Prompt: Let AI act as a ruthless reviewer Use case:

You have already generated one version of an image, but you cannot tell what is wrong with it.

Do not ask AI whether it “looks good.”

Instead, ask it to critique the image from the perspective of a Product Hunt user.

Prompt:

Please act as a Product Hunt gallery reviewer and critique this image for me.

Please give suggestions from the following angles:

1. Can users understand what the product does within 3 seconds?
2. Is the title clear?
3. Is the selling point obvious enough?
4. Is the layout too cluttered?
5. Is there too much text?
6. Is the product screenshot clear enough?
7. Does it make users want to see the next image?

Finally, please output:
- The top 3 biggest problems
- Specific improvement suggestions
- How the next image-generation prompt should be written

:::

Multiple AIs form a closed-loop workflow. They become a temporary design team, while I remain responsible for the final judgment and acceptance.

Step 6: Turn temporary work into a reusable workflow

For Draft’s Product Hunt Gallery, the process was roughly:

List all product features Gather the material.
Select the 3–4 features most suitable for the Gallery Do the subtraction.
Define one clear communication point for each image Clarify the purpose.
Find reference images on Product Hunt Find the visual direction.
Ask AI to analyze the reference images’ style, layout, and color palette Break down the design elements.
Create a unified style guide for the whole Gallery Define design rules before making images.
Make Feature images first, then the Hero image Work backwards.
Use a second AI as a reviewer for ruthless critique Bring in a third-party perspective.
Iterate based on feedback Do the final polishing.

If you also want to create a Product Hunt Gallery from scratch, you can follow this sequence directly.

Conclusion

My biggest takeaway is:

AI does not automatically make the result better.

The real value of AI is in breaking down tasks, building workflows, and making the process reusable.

AI can help preserve my thinking, edits, judgment, and collaboration process.

Product Hunt Gallery is just one example.

The same method can also be used for writing articles, making resumes, preparing interviews, designing courses, planning events, creating proposals, or any task you previously thought “requires a professional team.”

This is also why we built Draft.

Draft is a workspace for AI collaboration.

It helps organize your ideas, conversations, drafts, comments, tasks, and publishing process.

So they no longer get scattered.

So they can keep growing.

If you often use AI to write, build projects, or organize information, and you have started feeling that “my AI workflow is too scattered,” you are welcome to try Draft.